Professional Documents
Culture Documents
Sciencedirect: © 2015, Ifac (International Federation of Automatic Control) Hosting by Elsevier Ltd. All Rights Reserved
Sciencedirect: © 2015, Ifac (International Federation of Automatic Control) Hosting by Elsevier Ltd. All Rights Reserved
patients to operating rooms. The second is to provide each Job-shop problems called also Multipath workshops are
individual surgical unit responsible for synchronizing the workshops where operations are carried out in a fixed order,
various human and material resources used. vary according to the task. The flexible job-shop is an
extension of the classical job-shop model. Its uniqueness lies
We are particularly interested in human resources planning in
in the fact that several machines are potentially capable of
pediatric emergency organizations which is not yet processed
achieving a subset of operations. Basically, it is a problem of
according to the state of art analysis. This planning aims to
planning and organization of a set of tasks to be performed on
determine the best balance between patient demand and
a set of resources with variable performance (Gotha, 1993).
involved human resources in order to meet needs for treatment
Similarly, multi-skill health care tasks scheduling corresponds
while minimizing waiting times, optimizing the quality of
to assign health care tasks to medical staff members who are
health care services and reducing costs. This work belongs to
characterized by their skills and availabilities in order to satisfy
HOST project (Hospital: Optimization, Simulation and
patients’ needs while respecting their emergency degrees and
Crowding Avoidance) supported and financed by the French
taking into account their length of stay in the PED.
National Research Agency (ANR). It targets to optimize the
Pediatric Emergency Department (PED) functioning Given a set of tasks and a set of resources, resolving a
characterized by stochastic arrivals of patients which can lead scheduling problem corresponds to program tasks and allocate
to its overcrowding. It aims to better manage health care resources to optimize one or more goals (corresponding to
organizations, anticipate the overcrowding feature and objective performance criteria), respecting a set of constraints.
establish avoidance proposals for it. Our concerns focus on The problem of job-shop scheduling consists in organizing the
modeling the human resources allocation in the scheduling realization of N jobs on M machines and a job j represents a
process and also searching for the existence of an assignment number of nj non preemptable ordered operations. In multi-
allowing the realization of a health care plan. skill health care tasks scheduling, jobs corresponds to
treatment tasks (a treatment task for each patient), machines
This paper is organized as follows: an analogy between job-
are the medical staff members and health care operations being
shop scheduling and multi-skill health care tasks scheduling is
executed are also non preemptable. In both cases, the
presented in the second section followed by a mathematical
execution of each operation involves one resource selected
modelling of the studied problem. Then, the proposed method
from the set of available resources and at a given time, a
will be described in the fourth section. A scenario of
resource can only execute one operation: it becomes available
simulation is given in the fifth section. Finally, the last section
to other operations once the operation currently assigned is
is for conclusion and prospects.
accomplished (resource constraints). The assignment of an
2. ANALOGY BETWEEN JOB-SHOP SCHEDULING operation to a resource entails the occupancy of this resource
AND MULTI-SKILL HEALTH CARE TASKS during a processing time. So, for each flexible job-shop
SCHEDULING problem, we can associate a processing time to each operation.
However, because the most important target of health
Health care systems are similar to production systems which
organizations is to ensure a high quality of care, durations are
always try to meet demands (patients for the hospital and
not given importance while scheduling in health care process.
customers for industry). A health care system can be
In fact, durations of health care operations are uncertain and
considered as a full production system, constrained by limited
not known in advance. They depend on care providers’ skills
material and human capacity in order to deliver the best care and patients’ states. On the other hand, health care demands
at the lowest cost. Human resources planning is a central can occur unexpectedly. We cannot, in any circumstances
element of health institutions’ management because of the cost
know when the request arises. Requests are prioritized and
and the constraints related to it (skills and availability). The
scheduled according to emergency degrees. Some requests are
numerous hazards such as the stochastic arrivals of patients
less urgent than others; they can be delayed without
and the complications that can occur during treatment process, endangering patients’ lives. However, most of the requests in
the difficulty of standardization and the coordination of the PED and the emergency departments in general require
medical acts with the high number of actors make the planning
immediate intervention. According to the legal structure of the
a complex process. The implementation of generated plans
hospital, this latter has to accept all patients. In such a
requires sustained efforts.
situation, it is necessary to insert urgent patients in the
The search for industrial excellence in corporate comes near to planning already established, which sometimes causes
the concept of optimization of the quality of health care malfunctions and usually additional operating costs. This
facilities. The specific feature of health care systems is that makes the duration of a treatment operation, which is
they cannot speculate on human suffering and for which the determinist in job-shop scheduling problems, undergoes
objective of profit must be dismissed. significant variations depending on the type of operation, the
level of expertise of care provider, the patient and his
The most commonly discussed programming policies in pathology, etc. The variability of processing times often leads
previous works are those included in workshops. A workshop to changes in the schedule of the PED activity which may
is characterized by the number of machines it contains, its type cause a decline in quality concerning services provided to
and the order of their use to make a product. We distinguish patients. The consequences include long waiting times and
three types of workshops: flow-shop, job shop and open-shop. additional costs due to overtime. Hence, treatment tasks
We compare in this paper, the scheduling problem in the PED durations should be disintegrated while resolving human
to a job-shop scheduling problem in industry. resources affectation for health care operations execution. In
738
INCOM 2015
706 Sarah Ben Othman et al. / IFAC-PapersOnLine 48-3 (2015) 704–709
May 11-13, 2015. Ottawa, Canada
fact, durations of treatment tasks in emergency department are In this issue, we make the following assumptions:
difficult to calculate and cannot be known in advance. So we Each treatment task Tj can be started at the time t = rj;
chose to define the Demand Load (DL) of healthcare treatment Each medical staff member can execute at a time a single
in the PED to quantify patient treatment load. The more operation (resource constraint);
progression is, the less healthcare treatment demand will be. The total number of operations to be performed is higher
This reflects the progression of health care process for each than that of medical staff members.
patient and gives us an idea about health care operations
already executed and the ones which remain to be done. Performance Indicators
In job-shop scheduling a resource (machine) becomes To evaluate functioning quality of the PED, we define three
available to other tasks once the task which is currently performance indicators:
assigned to is completed. Task’s completion time can be Minimize the total load of all medical staff members:
calculated in advance. But, in health care treatment process, The total load is equal to the sum of the lengths of all health
the availabilities of medical staff members depend on their care operations executed according to any assignment.
skills and experience, patient’s health state, pathology severity However, for each operation Oi,j, the length of execution time
degree and the evolution of the current medical treatment task. is greater than the minimum length i, j where:
To set the availability date, it is hard for experts to afford exact i , j min (di , j ,k )
values due to the uncertainty involved. Besides, the evaluation k
is not the same in the eyes of the decision-makers (Issai and With i,j the minimum length of the health care operation
Singh, 2000), it depends on human feeling and recognition. So, execution for a treatment task Tj and Oi,j is the ith operation.
health care providers cannot make the single judgment
(Abbod, 2001). So, this criterion corresponds to minimize Cr1 i, j
j i
For the objectives to achieve, it is to minimize the overall Reduce waiting time of each patient
completion time (makespan), the total workload of the l
machines in job-shop scheduling, etc. However, in our study, It corresponds to minimize Cr2 = ∑j=1 max(0, cj − dj)
we aim also to minimize patients waiting time in addition to With:
balancing workload between all medical staff members and cj = the completion time of the treatment task Tj,
minimizing the response time as well as the workload of all the dj = the theoretical treatment time for the task Tj,
medical staff in the PED. l = the total number of treatment tasks.
3. MATHEMATICAL MODELING
Minimize response time to health care tasks
Notation
It corresponds to minimize Cr3 max r j i , j
M: number of available medical staff members, j
i
k: index of medical staff member Mk,
j: index of treatment task Tj, The proposed optimization method presented in this work will
nj : number of operations of treatment task Tj, focus on balancing the workload of medical staff members (k
rj : earliest starting date of task Tj, € [1…M]) as well as minimizing the response time for patients’
i: index of a health care operation, treatment.
Oi,j : ith operation of task Tj, 4. METHOD
ri,j : the earliest availability date of operation Oi,j,
The studied multi-skill health care tasks scheduling presents
Nt : total number of operations to execute, Nt n j ,
j two difficulties. The first one is to assign each operation Oi,j
Pt : total number of patients waiting for health care to a medical staff member Mk (selected from the set U). The
treatment Pt Pj , second one is the calculation of the starting times ti,j.
j The proposed method consists in two stages of resolution.
𝑑𝑑𝑖𝑖,𝑗𝑗,𝑘𝑘 : execution time of the operation Oi,j,
𝐶𝐶𝑖𝑖,𝑗𝑗,𝑘𝑘 : the skill of the medical staff member 𝑀𝑀k needed for 4.1 First stage
the execution of the operation 𝑂𝑂𝑖𝑖, 𝑗𝑗 ,
Assignment Algorithm
The problem is to organize the execution of N health care It allows us to assign each health care operation to the suitable
operations by M medical staff members. The set of medical medical staff member taking into account his availability date
staff members is U. Each task Tj corresponds to a patient and workloads of health care providers to whom operations
waiting for treatment in the PED and is a sequence of nj health have been already assigned. To compute the availability date
care operations. Each operation i of a task Tj (noted Oi,j) may of each MSk a simple application of fuzzy logic is proposed.
be performed by a set of medical staff members Ui,j U. Calculation is based on analysing the affordable skills, the
evolution of the current treatment task, patient’s health state
The affectation of an operation Oi,j to a medical staff member
and pathology severity degree. These are the inputs. Then, we
MSk Ui,j leads to the occupancy of this medical staff define for each input three sub-sets {“Low”, “Medium” and
member for a period di,j,k (we assume that di,j,k IN*). “High”}. Each subset is characterized by its trapezoidal
739
INCOM 2015
Sarah Ben Othman et al. / IFAC-PapersOnLine 48-3 (2015) 704–709 707
May 11-13, 2015. Ottawa, Canada
Membership Functions (MF) meanwhile the state varies representation since they act on binary genotypes. In a simple
gradually. After the definition of MF of the variables or genetic algorithm, the search is set by the successive
Inference which is based on decision rules depending on application of the variation operators. The cross is the phase of
experts’ views and historical data. cooperation between individuals while the mutation
corresponds to the individual adaptation phase.
Example of rules:
if (the medical staff is “high qualified”) and (the evolution of We consider the important characteristics of evolutionary
the current act is “high”) and (the pathology is “serious”) and algorithms and their relevance to solve NP-hard problems. We
(patient’s health state is “improves”) then (the medical staff is present some key points for solving approach:
“highly available”).
- A genetic representation (coding) appropriate to the problem
The result which is a fuzzy value undergoes a defuzzification
to determine possible solutions of the optimization problem;
to obtain an exact number as final output using the Center Of
Area method (COA). For the assignment, we choose to assign - Genetic operators that transform the composition of children
the health care operation to the medical staff member who during reproduction. Because a task must be treated by a single
corresponds to the highest fuzzy value which reflects his medical staff member selected from the set of members who
availability rate. If two medical staff members have the same are able to provide the corresponding health care service, we
fuzzy value, we make the choice while balancing the workload choose to correct the solutions generated by another operator
between all health care providers. to meet this requirement;
This assignment procedure allows us to construct a set E of
assignments (E = {Sz / 1≤z≤cardinal(E)}) and balance the - Parents are randomly selected from the current population for
medical staff members workload. Each assignment is the crossover and mutation with a probability of crossover p c
represented in a table Sz, Sz = {Szi,j,k / 1≤j≤N; 1≤i≤nj; 1≤k≤M}. (0 <pc <1) and a mutation probability pm (0 <pm <1);
For each i, j, k, the value of Szi,j,k can take 0 or 1. The value - We take a non-elitist replacement technique to build the new
population;
“Szi,j,k=1” means that Oi,j is assigned to MSk. The value “Szi,j,k=
- Fitness functions evaluate solutions based on two criteria:
0” means that Oi,j isn’t assigned to MSk. medical staff workload and response time.
Scheduling Algorithm In this stage, we generate from the set E constructed in the first
For each assignment, it calculates starting times ti,j by stage, an assignment scheme to control the genetic algorithm.
considering medical staff availabilities and precedence This scheme is going therefore to represent a constraint that
constraints. Conflicts are resolved by applying conventional must be respected by the new created individuals. This method
priority rules (SPT, LPT, FIFO, LIFO, FIRO (Boucon, 1991), consists in considering the assignments SZ given by the earlier
(Bel and Cavaille, 2001)), so we get a set of plans according to scheduling method and to determine (for each operation) the
the applied priority rules. In emergency department priority is set of possible medical staff members using a genetic
given at first to the most urgent cases, then to the patient who algorithm.
has arrived first. The scheduling procedure is as follows: Table 1. Medical Staff Skills
according to the availability date of a medical staff members
and the availability of the corresponding health care operation, C = {Ci,j,k / 1≤j≤N ; 1≤i≤nj ; 1≤k≤M}
the starting time of the operation is the minimum date among MS1 MS2 MS3 MS4
the two availabilities’ dates. O 1 ,1 0.2 0.6 0.5 0.2
T1 O 2 ,1 0.4 0.9 0.3 0.2
4.2 Second stage O 3 ,1 0.4 0.6 0.5 0.8
O 1 ,2 0.5 0.2 0.2 0.5
T2 O 2 ,2 0.3 0.4 1 0.4
The scheduling approach described in the previous paragraph O 3 ,2 1 0.2 0.3 0.3
can contribute to a multi-objective optimization by combining O 1 ,3 0.9 0.7 0.4 0.6
it with genetic algorithms and make develop an initial set of T3 O 2 ,3 0.5 0.6 0.9 0.2
solutions to a final one while improving the performance of
the whole system according to criteria we have fixed at the A possible scheduling plan related to medical staff skills is
beginning (Michalewicz, 1992), (Ono, 1996). given in Table 1. We consider that the assignment of a
health care operation Oi,j to a medical staff member is
Genetic algorithms are the most popular variant of possible when the competence Ci,j,k >=0.5.
evolutionary algorithms. Many specialists designate and
continue to designate the evolutionary approaches as "genetic Table 2. Assignment Sch
algorithms". As their name suggests, genetic algorithms are Sch = { Sch i,j,k / 1≤j≤N ; 1≤i≤nj ; 1≤k≤M}
based on the genetic inheritance of an individual (genotype)
MS1 MS2 MS3 MS4
represented by its chromosomes. The interaction of the O 1 ,1 0 * * 0
genotype of an individual with its environment determines its T1 O 2 ,1 0 1 0 0
phenotype which can be modified by mutation. Phenotype is O 3 ,1 0 * * *
evaluated by coding the genotype, which is often a binary O 1 ,2 * 0 0 *
symbol in order to provide a usable performance value by the T2 O 2 ,2 0 0 1 0
selection of operators. The variation operators (crossover and O 3 ,2 1 0 0 0
O 1 ,3 * * 0 *
mutation) presented above are related to the binary T3 O 2 ,3 * * * 0
740
INCOM 2015
708 Sarah Ben Othman et al. / IFAC-PapersOnLine 48-3 (2015) 704–709
May 11-13, 2015. Ottawa, Canada
The value " Schi,j,k = 0" indicates that the medical staff member Operator of mutation reducing the response time for
MSk is not enough qualified for this health care operation so treatment for patient j (RTLj)
we cannot assign it to him. Mutation 1
The value " Schi,j,k = 1" indicates that the assignment of the - Select randomly an individual S ;
operation Oi,j to the medical staff member MSk is obligatory - Choose the treatment task j whose response time the most long :
because he is the only one whose Ci,j,k > =0.5, in this case, all (Max j { RTLj such that RTLj =∑i∑k Si,j,k.d i,j,k }) ;
values of the rest of the line (i, j) are equal to "0". - i=1; r = 0 ;
- WHILE (i≤nj And r = 0)
The symbol: " * " indicates that the assignment is possible i.e. Find K0 such that Si,j,K0 =1;
(Ci,j,k >=0.5). FOR (k=1, k≤M)
We cannot have the value "1" and the symbol " * " in the same IF (d i,j,k < d i,j,k0 ) Then {Si,j,K0 =0; Si,j,K =1; r=1 ;}
line. End IF
End FOR
The application of the assignment procedure described in the i=i+1 ;
first stage may give as result the following scheduling S (Table End WHILE
3). - Call to "Scheduling_Algorithm" to calculate the starting and completion
times;
Table 3. Assignment S
The second mutation is responsible for the workload
S: possible scheduling
balancing:
MS1 MS2 MS3 MS4
O 1 ,1 0 0 1 0 Operator of mutation balancing workloads of medical staff
T1 O 2 ,1 0 1 0 0 members
O 3 ,1 0 1 0 0
O 1 ,2 0 0 0 1 Mutation 2
T2 O 2 ,2 0 0 1 0 - Select randomly an individual S ;
O 3 ,2 1 0 0 0 - Find the medical staff member who has the highest workload Mk1 (Maxk {
O 1 ,3 0 0 0 1 Wk / Wk=∑j∑i Si,j,k.di,j,k }) ;
T3 O 2 ,3 1 0 0 0 - Find the medical staff member who has the lowest workload M k2 (Min k {
Wk }) ;
a. Coding - Choose randomly an operation Oi, j such that S i ,j,k1 =1 ;
- Assign this operation to the medical staff member who has the lowest
workload: S i ,j,k1 =0 ; S i ,j,k2=1 ;
We represent the scheduling in the same assignment table S. - Calculate the starting and completion times according to the algorithm
Each Si,j,k=1 and Si,j,k = * are replaced by the couple ( ti,j, tfi,j ) "Scheduling_Algorithm";
where ti,j is the starting time and tfi,j is the completion time. Si,j,k
= 0 remain the same.
5. SIMULATION AND RESULTS
b. Crossover To better understand the proposed approach we present in this
section a scenario of a clinical case in the PED.
It consists in combining elements from two parent We suppose the arrival of 3 patients at time t = 0 to the health
chromosomes into one or more child chromosomes. care institution with 4 medical staff members mastering 3
(Michalewicz, 1992). It allows to create new combinations and types of skills, the degree is between 0 and 1.
enlarge our chance to find a better solution. Our operator uses
a Crossover Mask. We apply an efficient coding inspired from Table 4: Medical staff members
(Kacem et al., 2001) which respects our problem constraints. Medical Staff Description
MS1 Paediatrician
Cossover Algorithm MS2 Nurse1
1 2
- Select 2 parents S and S randomly; MS3 Nurse 2
- Select randomly 2 integers j and j’ such that j≤j’≤N ; MS4 Care assistant
- Select randomly 2 integers i and i' such that i≤nj and i’≤nj’ (in the case
where j=j', i≤i’) ;
- The assignment in f1 must match the same assignments in S1 for the set of
Patient 1 suffers from a mild concussion without loss of
operations between the line (i,j) and the (i’,j’); consciousness, patient 2 suffers from cardiopulmonary arrest
- The rest of assignments in f1 must match the same assignments in S2; and Patient 3 from an uncomplicated pneumonia.
- The assignment in f2 must match the same assignments in S2 for the set of The assignment table of medical staff members is given in the
operations between the line (i,j) and the (i’,j’);
- The rest of assignments in f2 must match the same assignments in S1;
table below:
- Call to "Scheduling_Algorithm" to calculate the starting and completion Table 5: Assignment table with medical staff skills
times;
MS1 MS2 MS3 MS4
O 1 ,1 0.9 0 0 0.9
c. Mutation T1 O 2 ,1 0 0 0.8 1
O 3 ,1 0.7 0 0.6 0
The genetic operators should be therefore able to ensure this O 1 ,2 0 1 1 0
T2 O 2 ,2 0.7 0.6 0 0
optimization according to the criteria presented in the previous
O 3 ,2 0 1 0.8 0
section. Thus, we propose two artificial mutation operators: O 1 ,3 0 0 0.6 0
the first one is responsible for reducing the response time for T3 O 2 ,3 0 0 0 1
patients’ treatment.
741
INCOM 2015
Sarah Ben Othman et al. / IFAC-PapersOnLine 48-3 (2015) 704–709 709
May 11-13, 2015. Ottawa, Canada
Table 6: Assignment using only the first step In our future works we will consider the rest of the
MS1 MS2 MS3 MS4
performance indicators and we will integrate our approach in
O 1 ,1 0 0 0 0, 10 an intelligent agent-based system.
T1 O 2 ,1 0 0 30, 50 0
REFERENCES
O 3 ,1 50, 80 0 0 0
O 1 ,2 0 0, 10 0 0 Abbod, M. F., Von Keyserlingk, D. G., Linkens, D. A. and
T2 O 2 ,2 0 10, 40 0 0 Mahfouf, M. (2001). Survey of utilization of fuzzy
O 3 ,2 0 0 0 40, 60
O 1 ,3 0 0 0, 30 0
technology in Medicine and Healthcare. Fuzzy Sets Syst.,
T3 O 2 ,3 0 0 0 30, 40 volume (120), N° 2, pp. 331-349.
Bel, G. and Cavaille, J-B. (2001). Production Scheduling, Chp
After applying the evolutionary approach, the final 6: Simulating Approach, under the supervision of Lopez,
assignment table is as follows: P. and Roubellat Hermes, F. France.
Table 7: Final assignment Bertrand, X. (2006). Press Kit, “Emergencies”, Ministry of
Health and Solidarity.
MS1 MS2 MS3 MS4 Boucon, D. (1991). Scheduling Workshop: assistance to the
O 1 ,1 0, 10 0 0 0 choice of priority rules. Ph.D. ENSAE, Toulouse, France.
T1 O 2 ,1 0 0 0 10, 20
O 3 ,1 20, 50 0 0 0
Canelon, M.A, Davilas, J. and Morles, E.C (2009). Intelligent
O 1 ,2 0 0, 10 0 0 supervision of petroleum processes based on multi-agent
T2 O 2 ,2 0 10, 40 0 0 systems. WSEAS TRANSACTIONS on SYSTEMS and
O 3 ,2 0 0 0 40, 60 CONTROL, ISSN: 1991-8763, Issue 9, volume (4), pp
O 1 ,3 0 0 0, 30 0 435-444.
T3 O 2 ,3 0 0 0 30, 40
Ghotha. (1993). Scheduling problems. Operational Research,
We notice that in table 7, the treatment task T1 has the longest volume (27), N° 1. pp 77-150. France.
treatment time. We have therefore reduce it to 50. Besides, the
Guinet, A. and Chaabane, A. (2003). Une approche de type
highest workload was 50, by applying the evolutionary MRP2 pour la gestion des blocs, Gestion et Ingénierie des
approach it became 40. The medical staff members have as a
Systèmes Hospitaliers GISEH’03, Luxembourg, France.
result balanced workload.
Isaai, M. T and Singh, M. G. (2000). An Objectif-oriented
Constraint-based Heuristic for a class of Passengers Train
100 Scheduling Problems, Journal of IEEE/SMC Part C:
80
60 Applications and Reviews, volume (30), n. 1, pp 12-21.
40 Jebali, A. and Ladet, P. (2003). Vers une affectation supervisée
20 dans les systèmes de production de soin, 1ère conférence
0 francophone en Gestion et Ingénierie des Systèmes
T1 T2 T3 MS1 MS2 MS3 MS4 Hospitaliers, GISEH’03, pp. 257-266, Lyon, France.
Response time Workload Kacem, I. Hammadi, S. and Borne, P. (2001). Direct
Chromosome Representation and Advanced Genetic
Using only step 1 Using the evolutionary approach
Operators for Job-shop Problems. Proceedings of the
International Conference on Computational Intelligence
Figure 1: Comparison between results of the first step for Modelling, Control and Automation (CIMCA'01), the
and results given by the evolutionary approach sepecial session on 'Multiobjective Evolutionary
Optimization', Las Vegas, USA.
Figure 1 shows that the evolutionary approach contributes to Kharraja. S. and Marcon, E. (2003). Vers une Construction
the improvement of our system performance. Balancing the automatique du Plan Directeur d’Allocation des Plages
workload between the health care providers leads to a less Horaires. 1ère Conférence Francophone en Gestion et
stressed medical staff and then to a higher performance and Ingénierie des Systèmes Hospitaliers, GISEH’03, Lyon,
response time minimization. France.
6. CONCLUSION Michalewicz, Z. (1992). Genetic Algorithms + Data Structures
= Evolution Programs. Springer Verlag.
In this paper, we have presented an efficient approach for
multi-skill health care tasks scheduling problems. Ono. (1996). A genetic algorithms for job-shop scheduling
This method evolves two steps: the first one is to apply the problems using job-based order crossover. In ICEC’96,
assignment procedure to solve the resource allocation problem Proceedings of the 1996 IEEE In conference on
and generate the assignment plans and the second one is to Evolutionary Computation, pp 574-552.
apply a controlled evolutionary algorithm. The initial Pham, M.C. (2002) Care of the patient in the hospital to better
population is constructed starting from the results of the first manage the complexity of the coordination of actors,
phase. This approach helps us to improve the performance of Dissertation of the National School of Public Health,
the proposed medical staff scheduling system. Renne.
742